Extended Data Fig. 1: Baseline demographics of cohort and aspects of sequencing coverage.

a, The demographic distributions of patient age, race, gender, mtDNA somatic mutation status, history of neoadjuvant treatment, mtDNA coverage, and tumor sample type for each of the cancer types included in our analysis. Somatic mutation status is annotated among the subset of samples with ≥90% paired tumor-normal mtDNA sequencing coverage (see Methods: classifying sample mtDNA variant status); mtDNA status distributions are shown for cancer types >10 such samples. Cancer types are ordered by increasing proportions of samples with VUS or truncating mtDNA mutations. b, Cancer type mtDNA coverage variation based on sequencing center. Center, the average percentage of mtDNA (among regions considered in our study) with sufficient coverage for calling mutations, compared between different cancer types in our cohort. Dot color indicates the sequencing center from which the exome sequencing data originate. Top, density histograms of the average % mtDNA coverage for each sequencing center. Samples sequenced at the Broad Institute are uniquely depleted for mtDNA off-target coverage. c, mtDNA coverage from off-target reads at each position. The number of samples for which the given mtDNA position was sequenced to at least 5 reads (top, the depth threshold used in our analyses) and 20 reads (bottom, for comparison). Red, the number of samples using unpaired tumor-only data, applicable only for protein-truncating variants which were always assumed to be of somatic origin; blue, the number using only matched-normal samples; green, the number of samples with coverage in both tumor and matched-normal samples at the given position (applicable for all non-truncating variants which required evidence that the variant was absent in the matched normal to be classified as somatic). Purple, the number of whole-genome sequenced samples available from ICGC/PCAWG for comparison. d, Proportion of samples with detectable mutations is not biased by cancer type sequencing coverage. There is no correlation between the fraction of well-covered samples in a cancer type and the proportion of well-covered samples with a detectable somatic mtDNA mutation. Cancer types with ≥30 well-covered samples shown, P-value and 95% confidence intervals from linear regression.